1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
6 #include <linux/swap.h>
8 #include <linux/blkdev.h>
10 #include <linux/iocontext.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/export.h>
15 #include <linux/mempool.h>
16 #include <linux/workqueue.h>
17 #include <linux/cgroup.h>
18 #include <linux/blk-cgroup.h>
19 #include <linux/highmem.h>
20 #include <linux/sched/sysctl.h>
21 #include <linux/blk-crypto.h>
22 #include <linux/xarray.h>
24 #include <trace/events/block.h>
26 #include "blk-rq-qos.h"
28 static struct biovec_slab
{
31 struct kmem_cache
*slab
;
32 } bvec_slabs
[] __read_mostly
= {
33 { .nr_vecs
= 16, .name
= "biovec-16" },
34 { .nr_vecs
= 64, .name
= "biovec-64" },
35 { .nr_vecs
= 128, .name
= "biovec-128" },
36 { .nr_vecs
= BIO_MAX_VECS
, .name
= "biovec-max" },
39 static struct biovec_slab
*biovec_slab(unsigned short nr_vecs
)
42 /* smaller bios use inline vecs */
44 return &bvec_slabs
[0];
46 return &bvec_slabs
[1];
48 return &bvec_slabs
[2];
49 case 129 ... BIO_MAX_VECS
:
50 return &bvec_slabs
[3];
58 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
59 * IO code that does not need private memory pools.
61 struct bio_set fs_bio_set
;
62 EXPORT_SYMBOL(fs_bio_set
);
65 * Our slab pool management
68 struct kmem_cache
*slab
;
69 unsigned int slab_ref
;
70 unsigned int slab_size
;
73 static DEFINE_MUTEX(bio_slab_lock
);
74 static DEFINE_XARRAY(bio_slabs
);
76 static struct bio_slab
*create_bio_slab(unsigned int size
)
78 struct bio_slab
*bslab
= kzalloc(sizeof(*bslab
), GFP_KERNEL
);
83 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", size
);
84 bslab
->slab
= kmem_cache_create(bslab
->name
, size
,
85 ARCH_KMALLOC_MINALIGN
, SLAB_HWCACHE_ALIGN
, NULL
);
90 bslab
->slab_size
= size
;
92 if (!xa_err(xa_store(&bio_slabs
, size
, bslab
, GFP_KERNEL
)))
95 kmem_cache_destroy(bslab
->slab
);
102 static inline unsigned int bs_bio_slab_size(struct bio_set
*bs
)
104 return bs
->front_pad
+ sizeof(struct bio
) + bs
->back_pad
;
107 static struct kmem_cache
*bio_find_or_create_slab(struct bio_set
*bs
)
109 unsigned int size
= bs_bio_slab_size(bs
);
110 struct bio_slab
*bslab
;
112 mutex_lock(&bio_slab_lock
);
113 bslab
= xa_load(&bio_slabs
, size
);
117 bslab
= create_bio_slab(size
);
118 mutex_unlock(&bio_slab_lock
);
125 static void bio_put_slab(struct bio_set
*bs
)
127 struct bio_slab
*bslab
= NULL
;
128 unsigned int slab_size
= bs_bio_slab_size(bs
);
130 mutex_lock(&bio_slab_lock
);
132 bslab
= xa_load(&bio_slabs
, slab_size
);
133 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
136 WARN_ON_ONCE(bslab
->slab
!= bs
->bio_slab
);
138 WARN_ON(!bslab
->slab_ref
);
140 if (--bslab
->slab_ref
)
143 xa_erase(&bio_slabs
, slab_size
);
145 kmem_cache_destroy(bslab
->slab
);
149 mutex_unlock(&bio_slab_lock
);
152 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned short nr_vecs
)
154 BIO_BUG_ON(nr_vecs
> BIO_MAX_VECS
);
156 if (nr_vecs
== BIO_MAX_VECS
)
157 mempool_free(bv
, pool
);
158 else if (nr_vecs
> BIO_INLINE_VECS
)
159 kmem_cache_free(biovec_slab(nr_vecs
)->slab
, bv
);
163 * Make the first allocation restricted and don't dump info on allocation
164 * failures, since we'll fall back to the mempool in case of failure.
166 static inline gfp_t
bvec_alloc_gfp(gfp_t gfp
)
168 return (gfp
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
)) |
169 __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
172 struct bio_vec
*bvec_alloc(mempool_t
*pool
, unsigned short *nr_vecs
,
175 struct biovec_slab
*bvs
= biovec_slab(*nr_vecs
);
177 if (WARN_ON_ONCE(!bvs
))
181 * Upgrade the nr_vecs request to take full advantage of the allocation.
182 * We also rely on this in the bvec_free path.
184 *nr_vecs
= bvs
->nr_vecs
;
187 * Try a slab allocation first for all smaller allocations. If that
188 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
189 * The mempool is sized to handle up to BIO_MAX_VECS entries.
191 if (*nr_vecs
< BIO_MAX_VECS
) {
194 bvl
= kmem_cache_alloc(bvs
->slab
, bvec_alloc_gfp(gfp_mask
));
195 if (likely(bvl
) || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
197 *nr_vecs
= BIO_MAX_VECS
;
200 return mempool_alloc(pool
, gfp_mask
);
203 void bio_uninit(struct bio
*bio
)
205 #ifdef CONFIG_BLK_CGROUP
207 blkg_put(bio
->bi_blkg
);
211 if (bio_integrity(bio
))
212 bio_integrity_free(bio
);
214 bio_crypt_free_ctx(bio
);
216 EXPORT_SYMBOL(bio_uninit
);
218 static void bio_free(struct bio
*bio
)
220 struct bio_set
*bs
= bio
->bi_pool
;
226 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, bio
->bi_max_vecs
);
229 * If we have front padding, adjust the bio pointer before freeing
234 mempool_free(p
, &bs
->bio_pool
);
236 /* Bio was allocated by bio_kmalloc() */
242 * Users of this function have their own bio allocation. Subsequently,
243 * they must remember to pair any call to bio_init() with bio_uninit()
244 * when IO has completed, or when the bio is released.
246 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
247 unsigned short max_vecs
)
249 memset(bio
, 0, sizeof(*bio
));
250 atomic_set(&bio
->__bi_remaining
, 1);
251 atomic_set(&bio
->__bi_cnt
, 1);
253 bio
->bi_io_vec
= table
;
254 bio
->bi_max_vecs
= max_vecs
;
256 EXPORT_SYMBOL(bio_init
);
259 * bio_reset - reinitialize a bio
263 * After calling bio_reset(), @bio will be in the same state as a freshly
264 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
265 * preserved are the ones that are initialized by bio_alloc_bioset(). See
266 * comment in struct bio.
268 void bio_reset(struct bio
*bio
)
271 memset(bio
, 0, BIO_RESET_BYTES
);
272 atomic_set(&bio
->__bi_remaining
, 1);
274 EXPORT_SYMBOL(bio_reset
);
276 static struct bio
*__bio_chain_endio(struct bio
*bio
)
278 struct bio
*parent
= bio
->bi_private
;
280 if (bio
->bi_status
&& !parent
->bi_status
)
281 parent
->bi_status
= bio
->bi_status
;
286 static void bio_chain_endio(struct bio
*bio
)
288 bio_endio(__bio_chain_endio(bio
));
292 * bio_chain - chain bio completions
293 * @bio: the target bio
294 * @parent: the parent bio of @bio
296 * The caller won't have a bi_end_io called when @bio completes - instead,
297 * @parent's bi_end_io won't be called until both @parent and @bio have
298 * completed; the chained bio will also be freed when it completes.
300 * The caller must not set bi_private or bi_end_io in @bio.
302 void bio_chain(struct bio
*bio
, struct bio
*parent
)
304 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
306 bio
->bi_private
= parent
;
307 bio
->bi_end_io
= bio_chain_endio
;
308 bio_inc_remaining(parent
);
310 EXPORT_SYMBOL(bio_chain
);
312 static void bio_alloc_rescue(struct work_struct
*work
)
314 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
318 spin_lock(&bs
->rescue_lock
);
319 bio
= bio_list_pop(&bs
->rescue_list
);
320 spin_unlock(&bs
->rescue_lock
);
325 submit_bio_noacct(bio
);
329 static void punt_bios_to_rescuer(struct bio_set
*bs
)
331 struct bio_list punt
, nopunt
;
334 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
337 * In order to guarantee forward progress we must punt only bios that
338 * were allocated from this bio_set; otherwise, if there was a bio on
339 * there for a stacking driver higher up in the stack, processing it
340 * could require allocating bios from this bio_set, and doing that from
341 * our own rescuer would be bad.
343 * Since bio lists are singly linked, pop them all instead of trying to
344 * remove from the middle of the list:
347 bio_list_init(&punt
);
348 bio_list_init(&nopunt
);
350 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
351 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
352 current
->bio_list
[0] = nopunt
;
354 bio_list_init(&nopunt
);
355 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
356 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
357 current
->bio_list
[1] = nopunt
;
359 spin_lock(&bs
->rescue_lock
);
360 bio_list_merge(&bs
->rescue_list
, &punt
);
361 spin_unlock(&bs
->rescue_lock
);
363 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
367 * bio_alloc_bioset - allocate a bio for I/O
368 * @gfp_mask: the GFP_* mask given to the slab allocator
369 * @nr_iovecs: number of iovecs to pre-allocate
370 * @bs: the bio_set to allocate from.
372 * Allocate a bio from the mempools in @bs.
374 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
375 * allocate a bio. This is due to the mempool guarantees. To make this work,
376 * callers must never allocate more than 1 bio at a time from the general pool.
377 * Callers that need to allocate more than 1 bio must always submit the
378 * previously allocated bio for IO before attempting to allocate a new one.
379 * Failure to do so can cause deadlocks under memory pressure.
381 * Note that when running under submit_bio_noacct() (i.e. any block driver),
382 * bios are not submitted until after you return - see the code in
383 * submit_bio_noacct() that converts recursion into iteration, to prevent
386 * This would normally mean allocating multiple bios under submit_bio_noacct()
387 * would be susceptible to deadlocks, but we have
388 * deadlock avoidance code that resubmits any blocked bios from a rescuer
391 * However, we do not guarantee forward progress for allocations from other
392 * mempools. Doing multiple allocations from the same mempool under
393 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
394 * for per bio allocations.
396 * Returns: Pointer to new bio on success, NULL on failure.
398 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned short nr_iovecs
,
401 gfp_t saved_gfp
= gfp_mask
;
405 /* should not use nobvec bioset for nr_iovecs > 0 */
406 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) && nr_iovecs
> 0))
410 * submit_bio_noacct() converts recursion to iteration; this means if
411 * we're running beneath it, any bios we allocate and submit will not be
412 * submitted (and thus freed) until after we return.
414 * This exposes us to a potential deadlock if we allocate multiple bios
415 * from the same bio_set() while running underneath submit_bio_noacct().
416 * If we were to allocate multiple bios (say a stacking block driver
417 * that was splitting bios), we would deadlock if we exhausted the
420 * We solve this, and guarantee forward progress, with a rescuer
421 * workqueue per bio_set. If we go to allocate and there are bios on
422 * current->bio_list, we first try the allocation without
423 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
424 * blocking to the rescuer workqueue before we retry with the original
427 if (current
->bio_list
&&
428 (!bio_list_empty(¤t
->bio_list
[0]) ||
429 !bio_list_empty(¤t
->bio_list
[1])) &&
430 bs
->rescue_workqueue
)
431 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
433 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
434 if (!p
&& gfp_mask
!= saved_gfp
) {
435 punt_bios_to_rescuer(bs
);
436 gfp_mask
= saved_gfp
;
437 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
442 bio
= p
+ bs
->front_pad
;
443 if (nr_iovecs
> BIO_INLINE_VECS
) {
444 struct bio_vec
*bvl
= NULL
;
446 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_iovecs
, gfp_mask
);
447 if (!bvl
&& gfp_mask
!= saved_gfp
) {
448 punt_bios_to_rescuer(bs
);
449 gfp_mask
= saved_gfp
;
450 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_iovecs
, gfp_mask
);
455 bio_init(bio
, bvl
, nr_iovecs
);
456 } else if (nr_iovecs
) {
457 bio_init(bio
, bio
->bi_inline_vecs
, BIO_INLINE_VECS
);
459 bio_init(bio
, NULL
, 0);
466 mempool_free(p
, &bs
->bio_pool
);
469 EXPORT_SYMBOL(bio_alloc_bioset
);
472 * bio_kmalloc - kmalloc a bio for I/O
473 * @gfp_mask: the GFP_* mask given to the slab allocator
474 * @nr_iovecs: number of iovecs to pre-allocate
476 * Use kmalloc to allocate and initialize a bio.
478 * Returns: Pointer to new bio on success, NULL on failure.
480 struct bio
*bio_kmalloc(gfp_t gfp_mask
, unsigned short nr_iovecs
)
484 if (nr_iovecs
> UIO_MAXIOV
)
487 bio
= kmalloc(struct_size(bio
, bi_inline_vecs
, nr_iovecs
), gfp_mask
);
490 bio_init(bio
, nr_iovecs
? bio
->bi_inline_vecs
: NULL
, nr_iovecs
);
494 EXPORT_SYMBOL(bio_kmalloc
);
496 void zero_fill_bio_iter(struct bio
*bio
, struct bvec_iter start
)
500 struct bvec_iter iter
;
502 __bio_for_each_segment(bv
, bio
, iter
, start
) {
503 char *data
= bvec_kmap_irq(&bv
, &flags
);
504 memset(data
, 0, bv
.bv_len
);
505 flush_dcache_page(bv
.bv_page
);
506 bvec_kunmap_irq(data
, &flags
);
509 EXPORT_SYMBOL(zero_fill_bio_iter
);
512 * bio_truncate - truncate the bio to small size of @new_size
513 * @bio: the bio to be truncated
514 * @new_size: new size for truncating the bio
517 * Truncate the bio to new size of @new_size. If bio_op(bio) is
518 * REQ_OP_READ, zero the truncated part. This function should only
519 * be used for handling corner cases, such as bio eod.
521 void bio_truncate(struct bio
*bio
, unsigned new_size
)
524 struct bvec_iter iter
;
525 unsigned int done
= 0;
526 bool truncated
= false;
528 if (new_size
>= bio
->bi_iter
.bi_size
)
531 if (bio_op(bio
) != REQ_OP_READ
)
534 bio_for_each_segment(bv
, bio
, iter
) {
535 if (done
+ bv
.bv_len
> new_size
) {
539 offset
= new_size
- done
;
542 zero_user(bv
.bv_page
, offset
, bv
.bv_len
- offset
);
550 * Don't touch bvec table here and make it really immutable, since
551 * fs bio user has to retrieve all pages via bio_for_each_segment_all
552 * in its .end_bio() callback.
554 * It is enough to truncate bio by updating .bi_size since we can make
555 * correct bvec with the updated .bi_size for drivers.
557 bio
->bi_iter
.bi_size
= new_size
;
561 * guard_bio_eod - truncate a BIO to fit the block device
562 * @bio: bio to truncate
564 * This allows us to do IO even on the odd last sectors of a device, even if the
565 * block size is some multiple of the physical sector size.
567 * We'll just truncate the bio to the size of the device, and clear the end of
568 * the buffer head manually. Truly out-of-range accesses will turn into actual
569 * I/O errors, this only handles the "we need to be able to do I/O at the final
572 void guard_bio_eod(struct bio
*bio
)
574 sector_t maxsector
= bdev_nr_sectors(bio
->bi_bdev
);
580 * If the *whole* IO is past the end of the device,
581 * let it through, and the IO layer will turn it into
584 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
587 maxsector
-= bio
->bi_iter
.bi_sector
;
588 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
591 bio_truncate(bio
, maxsector
<< 9);
595 * bio_put - release a reference to a bio
596 * @bio: bio to release reference to
599 * Put a reference to a &struct bio, either one you have gotten with
600 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
602 void bio_put(struct bio
*bio
)
604 if (!bio_flagged(bio
, BIO_REFFED
))
607 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
612 if (atomic_dec_and_test(&bio
->__bi_cnt
))
616 EXPORT_SYMBOL(bio_put
);
619 * __bio_clone_fast - clone a bio that shares the original bio's biovec
620 * @bio: destination bio
621 * @bio_src: bio to clone
623 * Clone a &bio. Caller will own the returned bio, but not
624 * the actual data it points to. Reference count of returned
627 * Caller must ensure that @bio_src is not freed before @bio.
629 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
631 WARN_ON_ONCE(bio
->bi_pool
&& bio
->bi_max_vecs
);
634 * most users will be overriding ->bi_bdev with a new target,
635 * so we don't set nor calculate new physical/hw segment counts here
637 bio
->bi_bdev
= bio_src
->bi_bdev
;
638 bio_set_flag(bio
, BIO_CLONED
);
639 if (bio_flagged(bio_src
, BIO_THROTTLED
))
640 bio_set_flag(bio
, BIO_THROTTLED
);
641 if (bio_flagged(bio_src
, BIO_REMAPPED
))
642 bio_set_flag(bio
, BIO_REMAPPED
);
643 bio
->bi_opf
= bio_src
->bi_opf
;
644 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
645 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
646 bio
->bi_iter
= bio_src
->bi_iter
;
647 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
649 bio_clone_blkg_association(bio
, bio_src
);
650 blkcg_bio_issue_init(bio
);
652 EXPORT_SYMBOL(__bio_clone_fast
);
655 * bio_clone_fast - clone a bio that shares the original bio's biovec
657 * @gfp_mask: allocation priority
658 * @bs: bio_set to allocate from
660 * Like __bio_clone_fast, only also allocates the returned bio
662 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
666 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
670 __bio_clone_fast(b
, bio
);
672 if (bio_crypt_clone(b
, bio
, gfp_mask
) < 0)
675 if (bio_integrity(bio
) &&
676 bio_integrity_clone(b
, bio
, gfp_mask
) < 0)
685 EXPORT_SYMBOL(bio_clone_fast
);
687 const char *bio_devname(struct bio
*bio
, char *buf
)
689 return bdevname(bio
->bi_bdev
, buf
);
691 EXPORT_SYMBOL(bio_devname
);
693 static inline bool page_is_mergeable(const struct bio_vec
*bv
,
694 struct page
*page
, unsigned int len
, unsigned int off
,
697 size_t bv_end
= bv
->bv_offset
+ bv
->bv_len
;
698 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) + bv_end
- 1;
699 phys_addr_t page_addr
= page_to_phys(page
);
701 if (vec_end_addr
+ 1 != page_addr
+ off
)
703 if (xen_domain() && !xen_biovec_phys_mergeable(bv
, page
))
706 *same_page
= ((vec_end_addr
& PAGE_MASK
) == page_addr
);
709 return (bv
->bv_page
+ bv_end
/ PAGE_SIZE
) == (page
+ off
/ PAGE_SIZE
);
713 * Try to merge a page into a segment, while obeying the hardware segment
714 * size limit. This is not for normal read/write bios, but for passthrough
715 * or Zone Append operations that we can't split.
717 static bool bio_try_merge_hw_seg(struct request_queue
*q
, struct bio
*bio
,
718 struct page
*page
, unsigned len
,
719 unsigned offset
, bool *same_page
)
721 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
722 unsigned long mask
= queue_segment_boundary(q
);
723 phys_addr_t addr1
= page_to_phys(bv
->bv_page
) + bv
->bv_offset
;
724 phys_addr_t addr2
= page_to_phys(page
) + offset
+ len
- 1;
726 if ((addr1
| mask
) != (addr2
| mask
))
728 if (bv
->bv_len
+ len
> queue_max_segment_size(q
))
730 return __bio_try_merge_page(bio
, page
, len
, offset
, same_page
);
734 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
735 * @q: the target queue
736 * @bio: destination bio
738 * @len: vec entry length
739 * @offset: vec entry offset
740 * @max_sectors: maximum number of sectors that can be added
741 * @same_page: return if the segment has been merged inside the same page
743 * Add a page to a bio while respecting the hardware max_sectors, max_segment
744 * and gap limitations.
746 int bio_add_hw_page(struct request_queue
*q
, struct bio
*bio
,
747 struct page
*page
, unsigned int len
, unsigned int offset
,
748 unsigned int max_sectors
, bool *same_page
)
750 struct bio_vec
*bvec
;
752 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
755 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
758 if (bio
->bi_vcnt
> 0) {
759 if (bio_try_merge_hw_seg(q
, bio
, page
, len
, offset
, same_page
))
763 * If the queue doesn't support SG gaps and adding this segment
764 * would create a gap, disallow it.
766 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
767 if (bvec_gap_to_prev(q
, bvec
, offset
))
771 if (bio_full(bio
, len
))
774 if (bio
->bi_vcnt
>= queue_max_segments(q
))
777 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
778 bvec
->bv_page
= page
;
780 bvec
->bv_offset
= offset
;
782 bio
->bi_iter
.bi_size
+= len
;
787 * bio_add_pc_page - attempt to add page to passthrough bio
788 * @q: the target queue
789 * @bio: destination bio
791 * @len: vec entry length
792 * @offset: vec entry offset
794 * Attempt to add a page to the bio_vec maplist. This can fail for a
795 * number of reasons, such as the bio being full or target block device
796 * limitations. The target block device must allow bio's up to PAGE_SIZE,
797 * so it is always possible to add a single page to an empty bio.
799 * This should only be used by passthrough bios.
801 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
802 struct page
*page
, unsigned int len
, unsigned int offset
)
804 bool same_page
= false;
805 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
806 queue_max_hw_sectors(q
), &same_page
);
808 EXPORT_SYMBOL(bio_add_pc_page
);
811 * bio_add_zone_append_page - attempt to add page to zone-append bio
812 * @bio: destination bio
814 * @len: vec entry length
815 * @offset: vec entry offset
817 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
818 * for a zone-append request. This can fail for a number of reasons, such as the
819 * bio being full or the target block device is not a zoned block device or
820 * other limitations of the target block device. The target block device must
821 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
824 * Returns: number of bytes added to the bio, or 0 in case of a failure.
826 int bio_add_zone_append_page(struct bio
*bio
, struct page
*page
,
827 unsigned int len
, unsigned int offset
)
829 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
830 bool same_page
= false;
832 if (WARN_ON_ONCE(bio_op(bio
) != REQ_OP_ZONE_APPEND
))
835 if (WARN_ON_ONCE(!blk_queue_is_zoned(q
)))
838 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
839 queue_max_zone_append_sectors(q
), &same_page
);
841 EXPORT_SYMBOL_GPL(bio_add_zone_append_page
);
844 * __bio_try_merge_page - try appending data to an existing bvec.
845 * @bio: destination bio
846 * @page: start page to add
847 * @len: length of the data to add
848 * @off: offset of the data relative to @page
849 * @same_page: return if the segment has been merged inside the same page
851 * Try to add the data at @page + @off to the last bvec of @bio. This is a
852 * useful optimisation for file systems with a block size smaller than the
855 * Warn if (@len, @off) crosses pages in case that @same_page is true.
857 * Return %true on success or %false on failure.
859 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
860 unsigned int len
, unsigned int off
, bool *same_page
)
862 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
865 if (bio
->bi_vcnt
> 0) {
866 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
868 if (page_is_mergeable(bv
, page
, len
, off
, same_page
)) {
869 if (bio
->bi_iter
.bi_size
> UINT_MAX
- len
) {
874 bio
->bi_iter
.bi_size
+= len
;
880 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
883 * __bio_add_page - add page(s) to a bio in a new segment
884 * @bio: destination bio
885 * @page: start page to add
886 * @len: length of the data to add, may cross pages
887 * @off: offset of the data relative to @page, may cross pages
889 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
890 * that @bio has space for another bvec.
892 void __bio_add_page(struct bio
*bio
, struct page
*page
,
893 unsigned int len
, unsigned int off
)
895 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
897 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
898 WARN_ON_ONCE(bio_full(bio
, len
));
904 bio
->bi_iter
.bi_size
+= len
;
907 if (!bio_flagged(bio
, BIO_WORKINGSET
) && unlikely(PageWorkingset(page
)))
908 bio_set_flag(bio
, BIO_WORKINGSET
);
910 EXPORT_SYMBOL_GPL(__bio_add_page
);
913 * bio_add_page - attempt to add page(s) to bio
914 * @bio: destination bio
915 * @page: start page to add
916 * @len: vec entry length, may cross pages
917 * @offset: vec entry offset relative to @page, may cross pages
919 * Attempt to add page(s) to the bio_vec maplist. This will only fail
920 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
922 int bio_add_page(struct bio
*bio
, struct page
*page
,
923 unsigned int len
, unsigned int offset
)
925 bool same_page
= false;
927 if (!__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
928 if (bio_full(bio
, len
))
930 __bio_add_page(bio
, page
, len
, offset
);
934 EXPORT_SYMBOL(bio_add_page
);
936 void bio_release_pages(struct bio
*bio
, bool mark_dirty
)
938 struct bvec_iter_all iter_all
;
939 struct bio_vec
*bvec
;
941 if (bio_flagged(bio
, BIO_NO_PAGE_REF
))
944 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
945 if (mark_dirty
&& !PageCompound(bvec
->bv_page
))
946 set_page_dirty_lock(bvec
->bv_page
);
947 put_page(bvec
->bv_page
);
950 EXPORT_SYMBOL_GPL(bio_release_pages
);
952 static void __bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
954 WARN_ON_ONCE(bio
->bi_max_vecs
);
956 bio
->bi_vcnt
= iter
->nr_segs
;
957 bio
->bi_io_vec
= (struct bio_vec
*)iter
->bvec
;
958 bio
->bi_iter
.bi_bvec_done
= iter
->iov_offset
;
959 bio
->bi_iter
.bi_size
= iter
->count
;
960 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
961 bio_set_flag(bio
, BIO_CLONED
);
964 static int bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
966 __bio_iov_bvec_set(bio
, iter
);
967 iov_iter_advance(iter
, iter
->count
);
971 static int bio_iov_bvec_set_append(struct bio
*bio
, struct iov_iter
*iter
)
973 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
974 struct iov_iter i
= *iter
;
976 iov_iter_truncate(&i
, queue_max_zone_append_sectors(q
) << 9);
977 __bio_iov_bvec_set(bio
, &i
);
978 iov_iter_advance(iter
, i
.count
);
982 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
985 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
986 * @bio: bio to add pages to
987 * @iter: iov iterator describing the region to be mapped
989 * Pins pages from *iter and appends them to @bio's bvec array. The
990 * pages will have to be released using put_page() when done.
991 * For multi-segment *iter, this function only adds pages from the
992 * next non-empty segment of the iov iterator.
994 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
996 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
997 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
998 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
999 struct page
**pages
= (struct page
**)bv
;
1000 bool same_page
= false;
1006 * Move page array up in the allocated memory for the bio vecs as far as
1007 * possible so that we can start filling biovecs from the beginning
1008 * without overwriting the temporary page array.
1010 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1011 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1013 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1014 if (unlikely(size
<= 0))
1015 return size
? size
: -EFAULT
;
1017 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1018 struct page
*page
= pages
[i
];
1020 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1022 if (__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1026 if (WARN_ON_ONCE(bio_full(bio
, len
)))
1028 __bio_add_page(bio
, page
, len
, offset
);
1033 iov_iter_advance(iter
, size
);
1037 static int __bio_iov_append_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1039 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1040 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1041 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
1042 unsigned int max_append_sectors
= queue_max_zone_append_sectors(q
);
1043 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1044 struct page
**pages
= (struct page
**)bv
;
1050 if (WARN_ON_ONCE(!max_append_sectors
))
1054 * Move page array up in the allocated memory for the bio vecs as far as
1055 * possible so that we can start filling biovecs from the beginning
1056 * without overwriting the temporary page array.
1058 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1059 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1061 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1062 if (unlikely(size
<= 0))
1063 return size
? size
: -EFAULT
;
1065 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1066 struct page
*page
= pages
[i
];
1067 bool same_page
= false;
1069 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1070 if (bio_add_hw_page(q
, bio
, page
, len
, offset
,
1071 max_append_sectors
, &same_page
) != len
) {
1080 iov_iter_advance(iter
, size
- left
);
1085 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1086 * @bio: bio to add pages to
1087 * @iter: iov iterator describing the region to be added
1089 * This takes either an iterator pointing to user memory, or one pointing to
1090 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1091 * map them into the kernel. On IO completion, the caller should put those
1092 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1093 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1094 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1095 * completed by a call to ->ki_complete() or returns with an error other than
1096 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1097 * on IO completion. If it isn't, then pages should be released.
1099 * The function tries, but does not guarantee, to pin as many pages as
1100 * fit into the bio, or are requested in @iter, whatever is smaller. If
1101 * MM encounters an error pinning the requested pages, it stops. Error
1102 * is returned only if 0 pages could be pinned.
1104 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1105 * responsible for setting BIO_WORKINGSET if necessary.
1107 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1111 if (iov_iter_is_bvec(iter
)) {
1112 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1113 return bio_iov_bvec_set_append(bio
, iter
);
1114 return bio_iov_bvec_set(bio
, iter
);
1118 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1119 ret
= __bio_iov_append_get_pages(bio
, iter
);
1121 ret
= __bio_iov_iter_get_pages(bio
, iter
);
1122 } while (!ret
&& iov_iter_count(iter
) && !bio_full(bio
, 0));
1124 /* don't account direct I/O as memory stall */
1125 bio_clear_flag(bio
, BIO_WORKINGSET
);
1126 return bio
->bi_vcnt
? 0 : ret
;
1128 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
1130 static void submit_bio_wait_endio(struct bio
*bio
)
1132 complete(bio
->bi_private
);
1136 * submit_bio_wait - submit a bio, and wait until it completes
1137 * @bio: The &struct bio which describes the I/O
1139 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1140 * bio_endio() on failure.
1142 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1143 * result in bio reference to be consumed. The caller must drop the reference
1146 int submit_bio_wait(struct bio
*bio
)
1148 DECLARE_COMPLETION_ONSTACK_MAP(done
,
1149 bio
->bi_bdev
->bd_disk
->lockdep_map
);
1150 unsigned long hang_check
;
1152 bio
->bi_private
= &done
;
1153 bio
->bi_end_io
= submit_bio_wait_endio
;
1154 bio
->bi_opf
|= REQ_SYNC
;
1157 /* Prevent hang_check timer from firing at us during very long I/O */
1158 hang_check
= sysctl_hung_task_timeout_secs
;
1160 while (!wait_for_completion_io_timeout(&done
,
1161 hang_check
* (HZ
/2)))
1164 wait_for_completion_io(&done
);
1166 return blk_status_to_errno(bio
->bi_status
);
1168 EXPORT_SYMBOL(submit_bio_wait
);
1171 * bio_advance - increment/complete a bio by some number of bytes
1172 * @bio: bio to advance
1173 * @bytes: number of bytes to complete
1175 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1176 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1177 * be updated on the last bvec as well.
1179 * @bio will then represent the remaining, uncompleted portion of the io.
1181 void bio_advance(struct bio
*bio
, unsigned bytes
)
1183 if (bio_integrity(bio
))
1184 bio_integrity_advance(bio
, bytes
);
1186 bio_crypt_advance(bio
, bytes
);
1187 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1189 EXPORT_SYMBOL(bio_advance
);
1191 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1192 struct bio
*src
, struct bvec_iter
*src_iter
)
1194 struct bio_vec src_bv
, dst_bv
;
1195 void *src_p
, *dst_p
;
1198 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1199 src_bv
= bio_iter_iovec(src
, *src_iter
);
1200 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1202 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1204 src_p
= kmap_atomic(src_bv
.bv_page
);
1205 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1207 memcpy(dst_p
+ dst_bv
.bv_offset
,
1208 src_p
+ src_bv
.bv_offset
,
1211 kunmap_atomic(dst_p
);
1212 kunmap_atomic(src_p
);
1214 flush_dcache_page(dst_bv
.bv_page
);
1216 bio_advance_iter_single(src
, src_iter
, bytes
);
1217 bio_advance_iter_single(dst
, dst_iter
, bytes
);
1220 EXPORT_SYMBOL(bio_copy_data_iter
);
1223 * bio_copy_data - copy contents of data buffers from one bio to another
1225 * @dst: destination bio
1227 * Stops when it reaches the end of either @src or @dst - that is, copies
1228 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1230 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1232 struct bvec_iter src_iter
= src
->bi_iter
;
1233 struct bvec_iter dst_iter
= dst
->bi_iter
;
1235 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1237 EXPORT_SYMBOL(bio_copy_data
);
1240 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1242 * @src: source bio list
1243 * @dst: destination bio list
1245 * Stops when it reaches the end of either the @src list or @dst list - that is,
1246 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1249 void bio_list_copy_data(struct bio
*dst
, struct bio
*src
)
1251 struct bvec_iter src_iter
= src
->bi_iter
;
1252 struct bvec_iter dst_iter
= dst
->bi_iter
;
1255 if (!src_iter
.bi_size
) {
1260 src_iter
= src
->bi_iter
;
1263 if (!dst_iter
.bi_size
) {
1268 dst_iter
= dst
->bi_iter
;
1271 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1274 EXPORT_SYMBOL(bio_list_copy_data
);
1276 void bio_free_pages(struct bio
*bio
)
1278 struct bio_vec
*bvec
;
1279 struct bvec_iter_all iter_all
;
1281 bio_for_each_segment_all(bvec
, bio
, iter_all
)
1282 __free_page(bvec
->bv_page
);
1284 EXPORT_SYMBOL(bio_free_pages
);
1287 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1288 * for performing direct-IO in BIOs.
1290 * The problem is that we cannot run set_page_dirty() from interrupt context
1291 * because the required locks are not interrupt-safe. So what we can do is to
1292 * mark the pages dirty _before_ performing IO. And in interrupt context,
1293 * check that the pages are still dirty. If so, fine. If not, redirty them
1294 * in process context.
1296 * We special-case compound pages here: normally this means reads into hugetlb
1297 * pages. The logic in here doesn't really work right for compound pages
1298 * because the VM does not uniformly chase down the head page in all cases.
1299 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1300 * handle them at all. So we skip compound pages here at an early stage.
1302 * Note that this code is very hard to test under normal circumstances because
1303 * direct-io pins the pages with get_user_pages(). This makes
1304 * is_page_cache_freeable return false, and the VM will not clean the pages.
1305 * But other code (eg, flusher threads) could clean the pages if they are mapped
1308 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1309 * deferred bio dirtying paths.
1313 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1315 void bio_set_pages_dirty(struct bio
*bio
)
1317 struct bio_vec
*bvec
;
1318 struct bvec_iter_all iter_all
;
1320 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1321 if (!PageCompound(bvec
->bv_page
))
1322 set_page_dirty_lock(bvec
->bv_page
);
1327 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1328 * If they are, then fine. If, however, some pages are clean then they must
1329 * have been written out during the direct-IO read. So we take another ref on
1330 * the BIO and re-dirty the pages in process context.
1332 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1333 * here on. It will run one put_page() against each page and will run one
1334 * bio_put() against the BIO.
1337 static void bio_dirty_fn(struct work_struct
*work
);
1339 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1340 static DEFINE_SPINLOCK(bio_dirty_lock
);
1341 static struct bio
*bio_dirty_list
;
1344 * This runs in process context
1346 static void bio_dirty_fn(struct work_struct
*work
)
1348 struct bio
*bio
, *next
;
1350 spin_lock_irq(&bio_dirty_lock
);
1351 next
= bio_dirty_list
;
1352 bio_dirty_list
= NULL
;
1353 spin_unlock_irq(&bio_dirty_lock
);
1355 while ((bio
= next
) != NULL
) {
1356 next
= bio
->bi_private
;
1358 bio_release_pages(bio
, true);
1363 void bio_check_pages_dirty(struct bio
*bio
)
1365 struct bio_vec
*bvec
;
1366 unsigned long flags
;
1367 struct bvec_iter_all iter_all
;
1369 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1370 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1374 bio_release_pages(bio
, false);
1378 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1379 bio
->bi_private
= bio_dirty_list
;
1380 bio_dirty_list
= bio
;
1381 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1382 schedule_work(&bio_dirty_work
);
1385 static inline bool bio_remaining_done(struct bio
*bio
)
1388 * If we're not chaining, then ->__bi_remaining is always 1 and
1389 * we always end io on the first invocation.
1391 if (!bio_flagged(bio
, BIO_CHAIN
))
1394 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1396 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1397 bio_clear_flag(bio
, BIO_CHAIN
);
1405 * bio_endio - end I/O on a bio
1409 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1410 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1411 * bio unless they own it and thus know that it has an end_io function.
1413 * bio_endio() can be called several times on a bio that has been chained
1414 * using bio_chain(). The ->bi_end_io() function will only be called the
1415 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1416 * generated if BIO_TRACE_COMPLETION is set.
1418 void bio_endio(struct bio
*bio
)
1421 if (!bio_remaining_done(bio
))
1423 if (!bio_integrity_endio(bio
))
1427 rq_qos_done_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
);
1430 * Need to have a real endio function for chained bios, otherwise
1431 * various corner cases will break (like stacking block devices that
1432 * save/restore bi_end_io) - however, we want to avoid unbounded
1433 * recursion and blowing the stack. Tail call optimization would
1434 * handle this, but compiling with frame pointers also disables
1435 * gcc's sibling call optimization.
1437 if (bio
->bi_end_io
== bio_chain_endio
) {
1438 bio
= __bio_chain_endio(bio
);
1442 if (bio
->bi_bdev
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1443 trace_block_bio_complete(bio
->bi_bdev
->bd_disk
->queue
, bio
);
1444 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1447 blk_throtl_bio_endio(bio
);
1448 /* release cgroup info */
1451 bio
->bi_end_io(bio
);
1453 EXPORT_SYMBOL(bio_endio
);
1456 * bio_split - split a bio
1457 * @bio: bio to split
1458 * @sectors: number of sectors to split from the front of @bio
1460 * @bs: bio set to allocate from
1462 * Allocates and returns a new bio which represents @sectors from the start of
1463 * @bio, and updates @bio to represent the remaining sectors.
1465 * Unless this is a discard request the newly allocated bio will point
1466 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1467 * neither @bio nor @bs are freed before the split bio.
1469 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1470 gfp_t gfp
, struct bio_set
*bs
)
1474 BUG_ON(sectors
<= 0);
1475 BUG_ON(sectors
>= bio_sectors(bio
));
1477 /* Zone append commands cannot be split */
1478 if (WARN_ON_ONCE(bio_op(bio
) == REQ_OP_ZONE_APPEND
))
1481 split
= bio_clone_fast(bio
, gfp
, bs
);
1485 split
->bi_iter
.bi_size
= sectors
<< 9;
1487 if (bio_integrity(split
))
1488 bio_integrity_trim(split
);
1490 bio_advance(bio
, split
->bi_iter
.bi_size
);
1492 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1493 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1497 EXPORT_SYMBOL(bio_split
);
1500 * bio_trim - trim a bio
1502 * @offset: number of sectors to trim from the front of @bio
1503 * @size: size we want to trim @bio to, in sectors
1505 void bio_trim(struct bio
*bio
, int offset
, int size
)
1507 /* 'bio' is a cloned bio which we need to trim to match
1508 * the given offset and size.
1512 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1515 bio_advance(bio
, offset
<< 9);
1516 bio
->bi_iter
.bi_size
= size
;
1518 if (bio_integrity(bio
))
1519 bio_integrity_trim(bio
);
1522 EXPORT_SYMBOL_GPL(bio_trim
);
1525 * create memory pools for biovec's in a bio_set.
1526 * use the global biovec slabs created for general use.
1528 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1530 struct biovec_slab
*bp
= bvec_slabs
+ ARRAY_SIZE(bvec_slabs
) - 1;
1532 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1536 * bioset_exit - exit a bioset initialized with bioset_init()
1538 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1541 void bioset_exit(struct bio_set
*bs
)
1543 if (bs
->rescue_workqueue
)
1544 destroy_workqueue(bs
->rescue_workqueue
);
1545 bs
->rescue_workqueue
= NULL
;
1547 mempool_exit(&bs
->bio_pool
);
1548 mempool_exit(&bs
->bvec_pool
);
1550 bioset_integrity_free(bs
);
1553 bs
->bio_slab
= NULL
;
1555 EXPORT_SYMBOL(bioset_exit
);
1558 * bioset_init - Initialize a bio_set
1559 * @bs: pool to initialize
1560 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1561 * @front_pad: Number of bytes to allocate in front of the returned bio
1562 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1563 * and %BIOSET_NEED_RESCUER
1566 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1567 * to ask for a number of bytes to be allocated in front of the bio.
1568 * Front pad allocation is useful for embedding the bio inside
1569 * another structure, to avoid allocating extra data to go with the bio.
1570 * Note that the bio must be embedded at the END of that structure always,
1571 * or things will break badly.
1572 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1573 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1574 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1575 * dispatch queued requests when the mempool runs out of space.
1578 int bioset_init(struct bio_set
*bs
,
1579 unsigned int pool_size
,
1580 unsigned int front_pad
,
1583 bs
->front_pad
= front_pad
;
1584 if (flags
& BIOSET_NEED_BVECS
)
1585 bs
->back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1589 spin_lock_init(&bs
->rescue_lock
);
1590 bio_list_init(&bs
->rescue_list
);
1591 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1593 bs
->bio_slab
= bio_find_or_create_slab(bs
);
1597 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1600 if ((flags
& BIOSET_NEED_BVECS
) &&
1601 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1604 if (!(flags
& BIOSET_NEED_RESCUER
))
1607 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1608 if (!bs
->rescue_workqueue
)
1616 EXPORT_SYMBOL(bioset_init
);
1619 * Initialize and setup a new bio_set, based on the settings from
1622 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
1627 if (src
->bvec_pool
.min_nr
)
1628 flags
|= BIOSET_NEED_BVECS
;
1629 if (src
->rescue_workqueue
)
1630 flags
|= BIOSET_NEED_RESCUER
;
1632 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
1634 EXPORT_SYMBOL(bioset_init_from_src
);
1636 static int __init
init_bio(void)
1640 bio_integrity_init();
1642 for (i
= 0; i
< ARRAY_SIZE(bvec_slabs
); i
++) {
1643 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1645 bvs
->slab
= kmem_cache_create(bvs
->name
,
1646 bvs
->nr_vecs
* sizeof(struct bio_vec
), 0,
1647 SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1650 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
1651 panic("bio: can't allocate bios\n");
1653 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
1654 panic("bio: can't create integrity pool\n");
1658 subsys_initcall(init_bio
);